Run 9. Outer Step 5. Inner Step 0. Peers 1.

config.json

@@ -1,135 +1,135 @@
 {
-  "_name_or_path": "manglofz/jung1",
+  "_name_or_path": "ytenszer/llamatrain4",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "NON_PARTICIPATING",
+    "102": "FAIL",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
+    "105": "FAIL",
     "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
-    "11": "FAIL",
-    "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
+    "109": "FAIL",
+    "11": "NON_PARTICIPATING",
+    "110": "NOT_ALIVE",
+    "111": "FAIL",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
     "117": "NON_PARTICIPATING",
-    "118": "FAIL",
+    "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "FAIL",
     "121": "NOT_ALIVE",
-    "122": "FAIL",
+    "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "NON_PARTICIPATING",
+    "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
     "127": "NON_PARTICIPATING",
     "128": "NOT_ALIVE",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "FAIL",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "FAIL",
     "136": "NOT_ALIVE",
-    "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "137": "NON_PARTICIPATING",
+    "138": "NOT_ALIVE",
     "139": "NOT_ALIVE",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
-    "143": "NON_PARTICIPATING",
+    "142": "FAIL",
+    "143": "NOT_ALIVE",
     "144": "NON_PARTICIPATING",
-    "145": "NON_PARTICIPATING",
-    "146": "NOT_ALIVE",
-    "147": "NON_PARTICIPATING",
+    "145": "NOT_ALIVE",
+    "146": "NON_PARTICIPATING",
+    "147": "NOT_ALIVE",
     "148": "NON_PARTICIPATING",
     "149": "FAIL",
-    "15": "NON_PARTICIPATING",
+    "15": "FAIL",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NON_PARTICIPATING",
     "154": "NON_PARTICIPATING",
     "155": "NON_PARTICIPATING",
-    "156": "FAIL",
-    "157": "NON_PARTICIPATING",
-    "158": "NOT_ALIVE",
+    "156": "NON_PARTICIPATING",
+    "157": "FAIL",
+    "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
-    "16": "NON_PARTICIPATING",
-    "160": "FAIL",
+    "16": "FAIL",
+    "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "FAIL",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
-    "167": "FAIL",
-    "168": "NON_PARTICIPATING",
-    "169": "FAIL",
+    "166": "NON_PARTICIPATING",
+    "167": "NOT_ALIVE",
+    "168": "FAIL",
+    "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
     "170": "FAIL",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
-    "175": "FAIL",
+    "173": "FAIL",
+    "174": "NON_PARTICIPATING",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "FAIL",
+    "177": "NON_PARTICIPATING",
     "178": "NOT_ALIVE",
     "179": "FAIL",
-    "18": "FAIL",
+    "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
-    "181": "NON_PARTICIPATING",
+    "181": "FAIL",
     "182": "NON_PARTICIPATING",
     "183": "FAIL",
-    "184": "NOT_ALIVE",
+    "184": "FAIL",
     "185": "FAIL",
-    "186": "FAIL",
+    "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
-    "189": "NOT_ALIVE",
-    "19": "FAIL",
-    "190": "FAIL",
+    "188": "FAIL",
+    "189": "NON_PARTICIPATING",
+    "19": "NOT_ALIVE",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "FAIL",
+    "193": "NOT_ALIVE",
     "194": "FAIL",
-    "195": "FAIL",
+    "195": "NON_PARTICIPATING",
     "196": "FAIL",
-    "197": "NON_PARTICIPATING",
-    "198": "FAIL",
+    "197": "FAIL",
+    "198": "NON_PARTICIPATING",
     "199": "FAIL",
-    "2": "FAIL",
-    "20": "NON_PARTICIPATING",
+    "2": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
-    "201": "NON_PARTICIPATING",
+    "201": "FAIL",
     "202": "NON_PARTICIPATING",
-    "203": "NOT_ALIVE",
+    "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "NON_PARTICIPATING",
-    "206": "FAIL",
+    "205": "FAIL",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
+    "209": "FAIL",
     "21": "NON_PARTICIPATING",
-    "210": "NON_PARTICIPATING",
-    "211": "FAIL",
-    "212": "NOT_ALIVE",
-    "213": "NON_PARTICIPATING",
+    "210": "FAIL",
+    "211": "NOT_ALIVE",
+    "212": "FAIL",
+    "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
@@ -137,126 +137,126 @@
     "218": "NON_PARTICIPATING",
     "219": "FAIL",
     "22": "NON_PARTICIPATING",
-    "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
-    "222": "FAIL",
-    "223": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
+    "221": "FAIL",
+    "222": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
+    "225": "NOT_ALIVE",
     "226": "NON_PARTICIPATING",
-    "227": "FAIL",
+    "227": "NOT_ALIVE",
     "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
+    "229": "NON_PARTICIPATING",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "FAIL",
+    "231": "NON_PARTICIPATING",
     "232": "NON_PARTICIPATING",
-    "233": "FAIL",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
-    "235": "FAIL",
+    "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
     "237": "FAIL",
     "238": "NON_PARTICIPATING",
     "239": "FAIL",
-    "24": "NON_PARTICIPATING",
-    "240": "NON_PARTICIPATING",
+    "24": "FAIL",
+    "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
-    "243": "NON_PARTICIPATING",
+    "243": "FAIL",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "FAIL",
-    "247": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "FAIL",
-    "251": "FAIL",
+    "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
     "26": "FAIL",
-    "27": "NON_PARTICIPATING",
+    "27": "FAIL",
     "28": "FAIL",
     "29": "NON_PARTICIPATING",
-    "3": "NOT_ALIVE",
-    "30": "NON_PARTICIPATING",
-    "31": "FAIL",
+    "3": "NON_PARTICIPATING",
+    "30": "FAIL",
+    "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "FAIL",
-    "35": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "FAIL",
     "37": "NOT_ALIVE",
     "38": "FAIL",
     "39": "NON_PARTICIPATING",
     "4": "FAIL",
-    "40": "NON_PARTICIPATING",
+    "40": "FAIL",
     "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
-    "43": "NOT_ALIVE",
+    "43": "FAIL",
     "44": "FAIL",
-    "45": "FAIL",
+    "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "FAIL",
+    "5": "NOT_ALIVE",
     "50": "FAIL",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
-    "53": "FAIL",
-    "54": "NOT_ALIVE",
+    "53": "NON_PARTICIPATING",
+    "54": "NON_PARTICIPATING",
     "55": "FAIL",
     "56": "NON_PARTICIPATING",
-    "57": "NOT_ALIVE",
-    "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
+    "57": "NON_PARTICIPATING",
+    "58": "NON_PARTICIPATING",
+    "59": "NON_PARTICIPATING",
     "6": "NON_PARTICIPATING",
-    "60": "NOT_ALIVE",
-    "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "60": "NON_PARTICIPATING",
+    "61": "FAIL",
+    "62": "NOT_ALIVE",
+    "63": "NOT_ALIVE",
     "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
-    "66": "NOT_ALIVE",
+    "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
     "7": "NON_PARTICIPATING",
     "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
-    "72": "FAIL",
+    "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
     "74": "NON_PARTICIPATING",
-    "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
+    "76": "FAIL",
+    "77": "NON_PARTICIPATING",
     "78": "FAIL",
-    "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
+    "79": "FAIL",
+    "8": "FAIL",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
-    "87": "NON_PARTICIPATING",
-    "88": "FAIL",
+    "86": "NON_PARTICIPATING",
+    "87": "FAIL",
+    "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NON_PARTICIPATING",
-    "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
+    "90": "FAIL",
+    "91": "NON_PARTICIPATING",
     "92": "NON_PARTICIPATING",
-    "93": "NON_PARTICIPATING",
-    "94": "FAIL",
-    "95": "NOT_ALIVE",
+    "93": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
+    "95": "NON_PARTICIPATING",
     "96": "NON_PARTICIPATING",
-    "97": "FAIL",
+    "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
-    "99": "FAIL"
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6450686,
+  "last_allreduce_block": 6451274,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae4d11ab369ff2253d312df96399b6398f4d07a64cd631fef330393ac97fed8b
+oid sha256:062e2b4d0ce1ac374c8f948a3143db39df5745cee2eedcf29d54c66219a8df57
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65c812a5899901bc3bf6b86614d05df0ed3b1c43ed61d9d0213e42de2f74fd40
+oid sha256:90ca19d200d958cab07922991bcf440fa63ba66022e9693651a1158064094ea5
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:298023f57fe32352daf1a4fadb6b54a55c0d3fdb1b5a9c2bf5cf15f2a85acda1
+oid sha256:8514f130f72f7d39a05543c095f71a1f2231cb56d7f6ac3e38b3588a55dce780
 size 4400252568